ANN-Based Inverse Modeling and Global Sensitivity Analysis of a CAR1 Damper

Friction dampers are widely used as passive energy dissipation devices in seismic protection systems; however, their response depends on numerous interacting parameters, complicating design. This study focuses on the CAR1 friction damper, investigating parameter influence and enabling efficient inverse identification of those that meet prescribed performance objectives. A finite element (FE) model reproduces the nonlinear damper behavior under seismic loading validated by previous experimental results. Based on the finite element (FE) dataset, an artificial neural network (ANN) is developed as a surrogate model to approximate the system response. This approach aims to overcome the excessive computational cost of the finite element method when performing optimization tasks involving numerous model evaluations. Global sensitivity analysis using the FAST and Sobol indices quantifies the influence of the parameters, revealing that a subset governs most of the variability for the target control axial force and dissipated energy. Building on these results, an inverse ANN has been contacted to optimize the parameters of the device based on (i) target control axial force, and (ii) maximum dissipated energy to a target displacement, providing a practical, physically informed tool for tailoring CAR1 friction dampers to specific seismic objectives.